What happens when you take AI, the primary focus of almost every business conversation today, and add quantum computing, itself a widely anticipated topic? The simple answer is fascinating enterprise potential, mixed enthusiasm, and, in many cases, quite a bit of confusion.
“There are some topics that people talk about so often as if they're the same thing,” says Scott Buchholz, managing director and head of quantum computing at Deloitte Consulting LLP.
So there's AI, machine learning, deep learning, quantum computing, and quantum AI and quantum machine learning, just to name a few terms that are often confused. All of these can be bandied about with little clarity in today's corporate environment.
What is Crystal Clear? We need to understand these and related innovations and start preparing, says Buchholz. “At the very least, leaders should spend time understanding quantum computing strategies,” he says. “Even if your strategy is to do nothing today, it's important to be cautious or you risk being caught unprepared.”
Entanglement conditions
Some disambiguation may be an important first step.
- machine learning has its roots in statistical models that evolved over time into neural networks. Neural networks form the basis of deep learning and enable much of the predictive analysis that businesses rely on today. “Machine learning allows us to create algorithms that can extrapolate from data we've seen before and apply those insights to new data today,” Buchholz explains. A good example is when trying to detect fraudulent credit card transactions. “Essentially, the algorithm tries to determine whether a particular transaction has characteristics in common with transactions that are known to be fraudulent in the past.”
- Generation AI This is the basis for large-scale language models, a machine learning technique applied at an unprecedented scale. Therefore, “machine learning algorithms are also inside generative AI,” Buchholz points out.
- “AI” in a broad sense It's even more difficult to pin down because it's been evolving since the 1960s, with definitions changing every few years. “People often say, and I'm not kidding, that 'AI' is just the frontier of what computers can't do yet,” Buchholz says.
Now add quantum to the mix.
- quantum computing Quantum computers operate differently than the classical computers in use today because they are based on qubits, which are quantum versions of classical binary bits. Quantum computing is expected to enable both opportunities and risks for businesses. Commercially relevant use cases are expected to be realized within one to three years, Buchholz said.
- quantum machine learning “We are using quantum computers and quantum algorithms to improve existing machine learning models,” said Mekenna McGrew, quantum information lead at Deloitte Consulting LLP. “It can be used to improve the representation of time series analysis, allowing us to create better predictive algorithms, for example.”
- quantum AI “Quantum machine learning” is a broad term that includes not only quantum machine learning, but also the use of AI to improve quantum computers, such as applying machine learning models to enable faster error correction in quantum computing hardware, McGrew said.
here and now
Currently, Buchholz points out, quantum machine learning can only be used at a limited scale.
“People's imaginations can stretch to large-scale language models running on quantum computers, but their capabilities are currently much more modest,” he says. “In fact, some may regret the current reality. Your laptop will probably be able to run more sophisticated models than many quantum computers can run today.”
Current quantum computers used in research today can process models with approximately a few hundred parameters. Large language models typically use billions or trillions.
“There's still a difference of several orders of magnitude,” Buchholz said.
But as quantum hardware improves, there may be exciting possibilities. So, “the dream is to be able to run larger and more complex machine learning models on quantum computers,” Buchholz says. “Quantum machine learning algorithms appear to be able to be trained to higher accuracy with less data than traditional algorithms.” These improvements could be useful in areas such as fraud and anomaly detection and predictive modeling for future predictions.
Some types of machine learning, such as those used to generate synthetic data, are likely to work particularly well on quantum computers, he added.
make the transition
Examples of potential applications for quantum computing are often focused on financial services and life sciences, but it could potentially be applied more broadly across industries, said Natasha Buckley, senior research leader for emerging issues at Deloitte's Center for Integrative Research and Deloitte Services LP.
No matter your industry, the disruptive potential of technology can motivate you to get started. Scenario planning helps companies anticipate and plan for different possible futures, rather than reacting reactively when the time comes, adds Buckley.
In the meantime, there are some practical ways to start your quantum journey and realize business value today by drawing inspiration from quantum computers while running traditional devices. “Today's servers can be used to run things like quantum computers, which can actually help today's machine learning and other tasks run more efficiently,” Buchholz explains.
This is known as “quantum-inspired” machine learning, and it “could have a huge impact,” McGrew said. “Quantum-inspired algorithms can enhance existing machine learning models in much the same way quantum machine learning promises, but can be scaled to full production with today's hardware.”
Quantum-inspired value
Quantum-inspired machine learning can run on the traditional server hardware that many companies already use to store data, whether on-premises or in a cloud environment. By experimenting with this, organizations can start building quantum skills and work toward strategic buy-in by demonstrating value early.
One potential use could be fraud detection, McGrew noted. “Quantum-inspired techniques can find fraudulent patterns that traditional machine learning might miss,” she says. “In use cases like fraud detection, incremental improvements in model quality can lead to large amounts of money.”
It's important to balance both quantum-inspired and quantum computing efforts, McGrew added. “Quantum computing use cases have the potential for long-term benefits,” she says. “A quantum-inspired approach will satisfy senior stakeholders today by delivering transformative value quickly.”
-by Katherine NoyesDeloitte Services LP, Senior Writer, Executive Perspectives, Wall Street Journal
